Projects

### Overview: Univer is an open-source SDK designed for creating productivity experiences within your product. It allows developers to embed spreadsheet, document, and presentation functionalities without being constrained by a hosted app or fixed UI. ### Key Features: - **Office Harness**: Supports spreadsheets, documents, presentations, bases, boards, and PDFs. - **Plugin Architecture**: Enables developers to create custom plugins for additional features. - **Canvas-Based Rendering**: Renders content in a customizable and interactive environment. - **Formula Engine**: Supports complex formulas and calculations. - **Facade API**: Provides a single API facade that works both in the browser and Node.js. - **Embeddable SDK**: Integrates seamlessly into existing applications. ### Tech Stack: TypeScript ### Why it Matters: Univer offers a flexible and customizable solution for building productivity tools, allowing developers to create unique and powerful applications without the limitations of a hosted environment. Its plugin architecture and open-source nature make it a valuable tool for both open-source and commercial projects alike.

### Overview OpenUI allows developers to describe and visualize user interfaces using their imagination, then render them in real-time. It simplifies the process of building UI components, making it fun, fast, and flexible. ### Key Features - Real-time rendering of UI components - Supports multiple frameworks: React, Svelte, Web Components - Live demo available at [try the demo](https://openui.fly.dev) - Environment variables for alternative models - Docker and Python scripts for easy deployment ### Tech Stack TypeScript ### Why It Matters OpenUI streamlines the UI development process, making it more enjoyable and efficient. It's particularly useful for developers working on complex applications that require dynamic UI components.

Overview: The repository is a collection of open-source implementations of various algorithms in C, licensed under the GPLv3, aimed at providing educational resources for students and educators. Key features: - Implementations of algorithms in C - Well-documented source code with detailed explanations - Atomic source codes using standard C library - Compiled and tested for MacOS and Ubuntu - Adherence to C11 standard for portability - Self-checks ensure correct implementations - Modular and open-source licensing for convenience Tech stack: C Why it matters: The repository offers a comprehensive learning resource for studying algorithms in a fundamental programming language, providing a solid foundation for further studies in computer science, mathematics, and data science.

Overview: This repository is a curated collection of AI engineering resources, covering a wide range of topics such as Large Language Models (LLMs), Retrieval-Augmented Generation (RAG), AI Agents, Model Carding, fine-tuning, evaluations, and production AI systems. It serves as a valuable resource for students and professionals looking to enhance their skills in AI engineering. Key features: - Comprehensive AI engineering resources - Detailed information on LLMs, RAG, AI Agents, Model Carding, fine-tuning, evaluations, and production systems - Useful for both students and professionals Tech stack: Python, JavaScript, HTML/CSS, Markdown, and various open-source tools Why it matters: This repository is essential for anyone interested in AI engineering, providing a wealth of resources and insights to help them develop and improve their skills in the field. Repository: dipakkr/awesome-ai-engineering GitHub stars: 22,411

### Overview: Video2X is a machine learning-based video super resolution and frame interpolation framework that has been updated to version 6.0.0. It offers improved performance, enhanced quality, and cross-platform support, making it a valuable tool for video processing enthusiasts and professionals. ### Key Features: - **Version 6.0.0 Highlights:** - Complete rewrite in C/C++ - Faster and more efficient architecture - Cross-platform support for Windows and Linux - Significant improvements in output quality - New GUI and installer for easy setup on Windows - Supports Anime4K v4, Real-ESRGAN, Real-CUGAN, and RIFE - Supports two modes: filtering (upscaling) and frame interpolation - Supports Anime4K v4 and all custom MPV-compatible GLSL shaders - Supports Real-ESRGAN, Real-CUGAN, and RIFE via ncnn and Vulkan - No additional disk space required during processing ### Tech Stack: - **Programming Language:** C++ - **GitHub Stars:** 22,289 ### Why It Matters: Video2X is a powerful tool for enhancing video quality, making it ideal for professionals and enthusiasts looking to improve their video editing and viewing experience. Its improvements in performance and quality make it a must-have for anyone involved in video processing and analysis. ### Hardware Requirements: - **CPU:** Haswell (Q2 2013) or newer - **GPU:** Vulkan support (Kepler (GTX

Overview: 这个项目提供了一个详细的100天机器学习学习路线，涵盖从零基础到进阶的全面内容，包括数据预处理、模型评价、现代机器学习库的使用方法等。 Key features: - 100天学习路线 - 中文讲解与练习 - 从Day 1～54覆盖课程 - 理论日提供本地说明和辅助资料 Tech stack: Python Why it matters: 为希望系统学习机器学习的开发者提供了一个全面且易于理解的100天学习路线，适合零基础和已有基础的学习者。 Repository: MLEveryday/100-Days-Of-ML-Code Primary language: Python GitHub stars: 22239 GitHub description: 100-Days-Of-ML-Code中文版 README excerpt: ```bash python -m venv .venv source .venv/bin/activate python -m pip install -r requirements.txt python "Code/Day 1_Data_Preprocessing.py" ``` 脚本默认把图形保存在`outputs/`。打开 Notebook 时按顺序运行所有单元；完整说明见安装文档。深度学习依赖单独安装，基础课程不需要 GPU。

### Overview: This repository contains the foundational models and architectures for large-scale self-supervised pre-training across tasks, languages, and modalities. It includes libraries and models for developing new architectures focused on both stability and capability, as well as efficiency and transferability. ### Key Features: - **DeepNet**: A library for scaling Transformers to 1,000 layers and beyond. - **Foundation Transformers (Magneto)**: Models towards true general-purpose modeling across tasks and modalities. - **Length-Extrapolatable Transformer**: A Transformer that can be length-extrapolated. - **X-MoE**: A library for scalable and finetunable sparse Mixture-of-Experts (MoE). - **BitNet**: 1-bit Transformers for Large Language Models. - **RetNet**: Retentive Network: A Successor to Transformer for Large Language Models. - **LongNet**: Scaling Transformers to 1,000,000,000 Tokens. - **Kosmos-2.5**: A Multimodal Literate Model. - **Kosmos-2**: Grounding Multimodal Large Language Models to the World. - **Kosmos-1**: A Multimodal Large Language Model (MLLM). - **MetaLM**: Language Models are General-Purpose Interfaces. - **DeltaLM/mT6**: Encoder-decoder pre-training for language generation and translation for 100+ languages. - **MiniLM**: Small and fast pre-trained models for language understanding and generation. - **AdaLM

Overview: ONNX Runtime is a cross-platform, high-performance machine learning inference and training accelerator. Key features: - Cross-platform support for various hardware, drivers, and operating systems. - Compatible with different machine learning frameworks, including PyTorch, TensorFlow/Keras, scikit-learn, LightGBM, and XGBoost. - Optimized performance leveraging hardware accelerators and graph optimizations. - Compatible with ONNX models for seamless integration with various frameworks. Tech stack: C++ Why it matters: ONNX Runtime simplifies the deployment of machine learning models, enabling faster customer experiences and lower costs through optimized performance and compatibility with diverse hardware and frameworks.

Overview: 🤗 Datasets is a comprehensive library for loading, processing, and manipulating datasets across various formats and modalities. Key features: - 📦 One-line dataloaders for major public datasets - 🔍 Multiple formats support - 🖼️ Multi-modal data handling - 🚀 Streaming mode for efficient data access Tech stack: Python Why it matters: 🤗 Datasets simplifies the process of working with large datasets, making it easier for developers to integrate them into their projects quickly and efficiently.

Overview: Docmost is an open-source alternative to Confluence and Notion, offering real-time collaboration, diagrams, spaces, permissions management, and more. Key features: - Real-time collaboration - Diagrams (Draw.io, Excalidraw, and Mermaid) - Spaces - Permissions management - Groups - Comments - Page history - Search - File attachments - Embeds (Airtable, Loom, Miro, and more) - Translations (10+ languages) Tech stack: TypeScript Why it matters: Docmost provides a robust and collaborative solution for documentation and wiki management, offering features that are both open-source and enterprise-ready, making it a valuable tool for developers and teams looking to enhance their project documentation and communication. Repository: docmost/docmost Primary language: TypeScript GitHub stars: 22,013 GitHub description: Docmost is an open-source collaborative wiki and documentation software. It is an open-source alternative to Confluence and Notion.

### Overview: ScreenPipe is a powerful tool designed to help developers automate repetitive tasks and streamline workflows in their company. By continuously recording computer activities, it maps out valuable work patterns and creates context for agents to perform tasks based on what you do. This approach ensures that agents can work efficiently without the need for interviews or consultants. ### Key Features: - **Record and Playback**: ScreenPipe captures every action you take on your computer, including screenshots, audio, and keystrokes. - **Workflow Mapping**: It generates a list of workflows that can be automated, reducing the need for manual interviews or hiring consultants. - **Agent Context**: Agents can be created based on your activities, making them highly context-aware and efficient. - **Search with AI**: You can search for anything using natural language, enhancing your productivity. - **Local Storage**: The captured history stays on your device by default, ensuring privacy and security. - **Source-Accessible**: You can inspect, modify, and audit the data, with a separate data path for cloud AI, sync, integrations, and managed team storage. ### Tech Stack: Rust ### Why It Matters: ScreenPipe is a game-changer for developers looking to automate repetitive tasks and streamline their workflows. By capturing and mapping your computer activities, it helps you find work worth automating and creates context for agents to perform tasks efficiently. This tool enhances productivity and reduces the need for manual interviews or consultants, making it an essential tool for modern development workflows.

Overview: Dyad is a local, open-source AI app builder that offers fast, private, and fully under your control functionality, similar to Lovable, v0, or Bolt, but running on your machine. Key features: - Local: Fast, private, and no lock-in. - Bring your own keys: Use your own AI API keys — no vendor lock-in. - Cross-platform: Easy to run on Mac or Windows. Tech stack: TypeScript Why it matters: Dyad provides a fast, private, and customizable alternative to Lovable, v0, and Bolt, offering developers a powerful tool for building AI applications without vendor lock-in.

Overview: Recommenders is a project that provides examples and best practices for building recommendation systems, including preparing data, building models, evaluating algorithms, tuning hyperparameters, and operationalizing models in a production environment. Key features: - Prepare Data (examples/01_prepare_data) - Model (examples/00_quick_start) - Evaluate (examples/03_evaluate) - Model Select and Optimize (examples/04_model_select_and_optimize) - Operationalize (examples/05_operationalize) Tech stack: Python Why it matters: Recommenders offers a comprehensive set of examples and best practices for researchers, developers, and enthusiasts in the field of recommendation systems, providing a practical guide for building, evaluating, and operationalizing recommendation models.

### Overview: 使用 dsh 插件，无需登录、注册或填写 API Key，即可访问包括 DeepSeek V4.1 Flash 和 Kimi K3 等前沿模型，完全免费且无使用限制。 ### Key features: - **无依赖安装**：只需在 dsh 中安装插件，无需任何配置。 - **免费模型**：涵盖 DeepSeek V4.1 Flash、Kimi K3 等前沿模型，无需注册、登录或填写 API Key。 - **实时更新**：模型清单跟随上游刷新，确保模型可用性。 - **无缓存**：插件不保存静态快照，模型列表随时间变化。 - **透明请求处理**：请求处理和数据流向由上游公开透明。 - **选择器优化**：网关拒绝的模型从下拉框移除，仅保留设置页记录。 - **公告中心**：仓库维护者编辑 JSON 并推送后，所有实例立即收到更新。 - **热重载**：升级与代码变更即时生效，无需重启应用。 - **响应体形状判定**：网关以 SSE 帧序列返回响应，插件按响应体形状判定。 - **思考强度**：模型档位菜单和预算阶梯根据模型声明和网关下发。 - **不依赖浏览器界面**：仅依赖 llm，不阻塞模型车道或影响设置页。 - **用量看板**：热力图、曲线等数据全部留在本地，不上传任何数据。 -

### Overview: Qwen (通义千问) is an advanced AI language model developed by Alibaba Cloud. It is designed for high-performance chat interactions and pretraining tasks. The model is based on the Qwen architecture, which has been trained on a vast corpus of text data to achieve exceptional performance in language understanding and generation. ### Key Features: - **Model Size:** - Qwen-1.8B: 1.8B parameters - Qwen-7B: 7B parameters - Qwen-14B: 14B parameters - Qwen-72B: 72B parameters - **Int8 and Int4 Versions:** - Qwen-1.8B (Int8): 1.8B parameters with integer 8-bit precision - Qwen-1.8B (Int4): 1.8B parameters with integer 4-bit precision - Qwen-7B (Int8): 7B parameters with integer 8-bit precision - Qwen-7B (Int4): 7B parameters with integer 4-bit precision - Qwen-14B (Int8): 14B parameters with integer 8-bit precision - Qwen-14B (Int4): 14B parameters with integer 4-bit precision - Qwen-72B (Int8): 72B parameters with integer 8-bit precision - Qwen-7

你好！很高兴为你服务。有什么我能帮助你的吗？

### Overview: Agent Development Kit (ADK) is an open-source, code-first Python framework designed for building, evaluating, and deploying sophisticated AI agents with flexibility and control. It simplifies the process of creating, deploying, and orchestrating agent workflows, making it an essential tool for developers looking to harness the power of AI in their projects. ### Key Features: - **Workflow Runtime**: A graph-based execution engine for composing deterministic execution flows for agentic apps, supporting routing, fan-out/fan-in, loops, retry, state management, dynamic nodes, human-in-the-loop, and nested workflows. - **Task API**: Structured agent-to-agent delegation with multi-turn task mode, single-turn controlled output, mixed delegation patterns, human-in-the-loop, and task agents as workflow nodes. - **Modular Multi-Agent Systems**: Design scalable applications by composing multiple specialized agents into flexible hierarchies. - **Rich Tool Ecosystem**: Utilize pre-built tools, custom functions, OpenAPI specs, MCP tools, or integrate existing tools to give agents diverse capabilities, all for tight integration with the Google ecosystem. - **Code-First Development**: Define agent logic, tools, and orchestration directly in Python for ultimate flexibility, testability, and versioning. - **Agent Config**: Build agents without code. Check out the [Agent Config](https://adk.dev/agents/config/) feature. - **Tool Confirmation**: A [tool confirmation flow (HITL)](https://adk.dev/tools-custom/confirmation/) that can guard tool execution with explicit confirmation and custom input.

Overview: Jina-Serve is a framework designed for building and deploying AI services that communicate via gRPC, HTTP, and WebSockets. It supports multiple ML frameworks and data types, providing high-performance service design and seamless scaling. Key features: - Native support for major ML frameworks and data types - High-performance service design with scaling, streaming, and dynamic batching - LLM serving with streaming output - Built-in Docker integration and Executor Hub - One-click deployment to Jina AI Cloud - Enterprise-ready with Kubernetes and Docker Compose support Tech stack: Python Why it matters: Jina-Serve simplifies the process of building and deploying AI services, making it easier to scale and deploy services efficiently, while also providing a robust and scalable solution for various ML tasks. GitHub stars: 21854

Overview: PEFT stands for Parameter-Efficient Fine-Tuning, enabling efficient adaptation of large pretrained models to various downstream applications by fine-tuning only a small number of parameters instead of all of them. This significantly reduces computational and storage costs. Key features: - 📦 PEFT library: A comprehensive Python library for PEFT methods. - 📦 Transformers integration: Easy model training and inference with Transformers. - 📦 Diffusers: Convenient management of adapters. - 📦 Accelerate: Distributed training and inference for large models. Tech stack: Python Why it matters: PEFT methods are crucial for making large pretrained models accessible and practical for a wide range of applications, especially in scenarios where computational resources are limited. Repository: huggingface/peft GitHub stars: 21,796

Overview: pytorch handbook 是一本开源书籍，旨在帮助希望使用 PyTorch 进行深度学习开发和研究的朋友快速入门。教程通过测试确保可以成功运行。 Key features: - PyTorch 环境搭建 - PyTorch 深度学习：60分钟快速入门 - 引用网上的资料，对每个引用附上原文链接 - 包含 PyTorch 的基础教程、实践教程、提高教程和应用教程 - 提供可视化工具（visdom、tensorboardx）和多GPU并行训练技术 Tech stack: Jupyter Notebook Why it matters: 本书通过提供详细的教程和实用的资源，帮助读者快速上手 PyTorch，同时保持与 PyTorch 最新版本的同步，确保教程内容的准确性和实用性。